Leopold Ashton-Bradner’s AI thesis survived July’s wipeout; the leverage did not. The fund reportedly fell about $35 billion in assets after a 67% down July, margin calls forced a Citadel unwind, and yet it still sits up roughly 80% YTD after returning 439% by end-June 2026 and more than 1,500% since inception. The core bet stays the same: more AI models, inference, compute, HBM, advanced chips, fabs, semiconductor equipment, data centers, electricity, then physical intelligence and robots. The biggest read-through is that the public market can still express the same bottlenecks through ASML, TSMC, MU, and SK Hynix, while the private-book names keep shifting toward Anthropic, Fluidstack, Mat X, Source Foundry, and a new robotics AI company.
A July blow-up reportedly cut Leopold Ashton-Bradner’s fund assets by about $35 billion after a very tough month; the damage came from forced deleveraging and selling the book to Citadel, not just mark-to-market losses.
Situational Awareness was running a barbell: long AI-buildout suppliers and short software names expected to be disrupted by AI, then layering leverage and options on top.
Wall Street Journal says the fund at times borrowed an extra 3-4x for every $1 of capital; the resulting leverage magnified July’s losses when AI infrastructure stocks fell and shorted software names rallied.
Ken Griffin negotiated through the night, bought the vast majority of Leopold’s public stock portfolio at a discount of more than 10% to market value, and Citadel’s stocks bounced immediately after the unwind.
Citadel’s equity fund gained roughly 14% in July and Wellington gained about 6% in July, showing how fast the sold names recovered after the forced sale.
By end-June 2026 the fund was reported up 439% on the year and more than 1,500% since inception; even after a 67% down July it was still up roughly 80% YTD.
The fund’s money keeps coming back: about a week after the blow-up, it started seeing a flood of inbound requests from investors wanting to give it more capital.
Public-market investors were pointed to the same bottlenecks as the private portfolio: ASML, TSMC, MU, and SK Hynix, with MU and SK Hynix described as very expensive.
The private portfolio’s biggest current position is Anthropic; Situational Awareness joined its May round that raised $65 billion at a $965 billion post-money valuation, and Anthropic has filed paperwork to eventually go public.
Fluidstack is another large bet, positioned as compute infrastructure for frontier AI labs that need land, electricity, cooling, networking, financing, and potentially gigawatts of power.
Mat X is a portfolio bet on specialized AI chips for large-language-model workloads; Situational Awareness co-led a $500 million financing round earlier this year.
Situational Awareness also anchored, with Lux Capital, an approximately $1 billion robotics AI company focused on foundation models for robots that can execute instructions like unloading a dishwasher or making a sandwich.
A mysterious new $400 million private investment went into Source Foundry, on top of an earlier $100 million, bringing the total to $500 million; the company, founded in 2025, is building tools for manufacturing advanced chips.
The thesis stack ties together Anthropic for intelligence, physical intelligence for robots, Fluidstack for training and inference infrastructure, Mat X for chips, and Source Foundry for semiconductor equipment; the macro chain runs toward more AI, more compute, more HBM, more fab capacity, more equipment, more data centers, and more electricity.
ASTC’s own timeline moved fast: TSA added Tracer 100 to its air cargo security technology list in June 2024, DHS gave First Warn Detect an R&D contract in January 2025, ASTC booked a $429,000 sale of six TRACER 1000 ETDs in April 2025, and in May 2026 the TRACER 1000 won ECAC/EU G1 approval and then ECAC certification for wand swabbing.
ASTC’s commercial pitch is a mass-spectrometry-based alternative to IMS ETDs: the Tracer 100 is said to have a virtually unlimited threat library, avoid false alarms from personal care products and household chemicals, and run in about 35-37 locations across 16 countries.
ASTC’s new lunar resource and infrastructure program is the big risk/reward on the sponsored side: the board approved it in May 2026, ASTC filed in June 2026 for about $20 million of NASA CLPS 2 non-dilutive funding, and the board also approved exploring a sale of First Detect even as the company remains unprofitable, modest-revenue, and potentially dilutive with a $200 million shelf on file.